Publication

Abstract : Emotion analysis often relies on annotated data sets, where labels assigned to documents represent the average or majority opinion of annotators. These models are effective in identifying general views but have limited ability to predict emotions experienced by individuals. In our study, we conducted research on a Kannada data set comprising 4000 opinions. Each opinion was annotated by a group of 15 individuals, providing assessments on two aspects: valence and arousal. Additionally, we analyzed the intensity of eight emotions based on Plutchik’s model. This investigation revealed significant variations in individual responses compared to the average. To address this, we introduced a new measure called Personal Emotional Bias (PEB) to estimate this effect. Incorporating PEB proved crucial in enhancing personalized reasoning. Our method and measure have the potential to enhance the effectiveness of individualized solutions and content suggestion systems designed to shield annotators from offensive and undesirable content that is inherently subjective.